How Do New Developers Onboard Faster with AI?
For New hires and junior developers onboarding to enterprise codebases · Based on Priscila Andre's Comprehend-First AI Coding Skill
// TL;DR
New developers can use Priscila Andre's Comprehend-First AI Coding Skill to onboard to unfamiliar enterprise codebases in days instead of weeks. By using the Catch Me Up prompt with Architecture, Convention, and Testing exploration modes, you build an accurate mental model of the codebase before writing your first line of code. This replaces the awkward 'reading code for a week' phase with structured, AI-accelerated comprehension that produces diagrams, component trees, and verified understanding.
Why Is Codebase Onboarding So Slow for New Developers?
Onboarding to a large, established codebase is one of the most frustrating experiences in software engineering. You're expected to contribute quickly, but the repository has years of history, undocumented conventions, and architectural decisions that only make sense if you were there when they were made.
Traditionally, new developers spend their first week or two reading code, asking questions when they can find a free senior engineer, and hoping their mental model is roughly correct before submitting their first PR.
Priscila Andre's Comprehend-First AI Coding Skill replaces this painful ramp-up with a structured, repeatable process. Instead of passively reading code, you actively interrogate it using AI as your always-available senior engineer.
How Do You Use Catch Me Up on Your First Day?
Start by creating a prompt using the Catch Me Up format:
1. Declare your role: "I am a new contributor to this repository. I have no prior context."
2. Select exploration modes: Begin with Architecture (to understand overall structure) and Convention (to learn coding standards).
3. Ask a specific question: "How is this repository structured, and what are the key naming conventions and lint rules I should follow?"
4. Request visual output: "Show me a component tree and a table of the main modules with their responsibilities."
The AI produces a structured summary you can internalize in minutes rather than days. But critically, you don't stop there—you verify. Ask follow-up questions: "Does module X actually communicate with module Y, or is there an intermediary?" This verification step is where real understanding is built.
What Exploration Modes Matter Most for Onboarding?
For your first few days, focus on three modes:
- Architecture: Gives you the structural overview—what components exist, how they relate, and where boundaries are.
- Convention: Reveals the coding standards, lint rules, and naming patterns you must follow. This prevents your first PR from being rejected for style violations.
- Testing: Shows you how tests are structured, what they actually simulate vs. intercept, and where you should add tests for your contributions.
Once you have these three foundations, use Feature mode to understand the specific area you'll be working in, and Trace mode to follow the execution path of the feature you're about to modify.
How Do You Know When You're Ready to Write Code?
Apply the comprehension gate from the Research → Comprehend → Plan → Implement workflow. Ask yourself: Can I explain, in plain language, what this code does and why? If yes, move to planning. If no, run another round of Catch Me Up with a more targeted question.
The goal is not to understand the entire codebase—it's to understand enough about the area you're working in to produce what Priscila Andre calls keynote code: intentional, well-understood contributions you'd be proud to present publicly.
Don't fall into the trap of skipping comprehension because you feel pressure to ship. Slop code—code shipped without understanding—creates more work for your team and undermines your credibility as a new contributor.
What Should You Do Next?
Create a Catch Me Up Markdown template and save it in your dotfiles or project root. Every time you encounter an unfamiliar part of the codebase, invoke it with the appropriate exploration mode. Within your first week, you'll have a structured, verified mental model that would normally take a month to build through passive code reading. Start with Architecture mode today.
// FREQUENTLY ASKED QUESTIONS
How long does it take to onboard to a new codebase using Comprehend-First?
Most developers can build a working mental model of their assigned area within 1–3 days using structured Catch Me Up prompts, compared to 1–3 weeks of passive code reading. The time depends on codebase complexity and how many exploration modes you need. Each focused session takes 15–30 minutes, and you can run several per day.
Should I still ask my team questions if I'm using AI for onboarding?
Yes, but AI handles the questions you'd hesitate to ask a busy colleague—basic architecture, naming conventions, test structure. Save your team's time for nuanced questions about design decisions and organizational context that aren't captured in code. AI is the cheapest senior engineer available; use it for the high-volume, low-stakes questions first.
What if the AI gives me wrong information about the codebase during onboarding?
This is expected and is part of the process. Always verify AI output against actual code, documentation, and commit history. When you find a discrepancy, correct the AI and re-analyze. These discrepancies often highlight poorly documented or counterintuitive areas of the codebase—flagging them early makes you a more valuable contributor.